CN106875381A - A kind of phone housing defect inspection method based on deep learning - Google Patents
A kind of phone housing defect inspection method based on deep learning Download PDFInfo
- Publication number
- CN106875381A CN106875381A CN201710034677.0A CN201710034677A CN106875381A CN 106875381 A CN106875381 A CN 106875381A CN 201710034677 A CN201710034677 A CN 201710034677A CN 106875381 A CN106875381 A CN 106875381A
- Authority
- CN
- China
- Prior art keywords
- image
- feature extraction
- phone housing
- feature
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 110
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 238000007689 inspection Methods 0.000 title claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 73
- 238000000605 extraction Methods 0.000 claims abstract description 73
- 238000012549 training Methods 0.000 claims abstract description 22
- 230000002950 deficient Effects 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000010606 normalization Methods 0.000 claims description 15
- 230000009466 transformation Effects 0.000 claims description 13
- 238000012937 correction Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 7
- 238000013519 translation Methods 0.000 claims description 7
- 238000005520 cutting process Methods 0.000 claims description 5
- 238000005286 illumination Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 abstract description 7
- 210000004027 cell Anatomy 0.000 description 19
- 230000000694 effects Effects 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/2163—Partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
- G06T3/147—Transformations for image registration, e.g. adjusting or mapping for alignment of images using affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/48—Extraction of image or video features by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of phone housing defect inspection method based on deep learning, the method comprises the following steps：(1) obtain phone housing image to be detected and pre-processed；(2) be input into pretreated image carries out defects detection and obtains the position of existing defects on phone housing to the good defects detection model of training in advance, and provides the confidence level that the position is defect；Wherein, defects detection model is the depth network based on deep learning, including the feature extraction network for cascading successively and grader and recurrence device network, described feature extraction network carries out feature extraction to the image for pre-processing and obtains characteristic image, and described grader carries out classification recurrence to characteristic image and obtains phone housing defective locations and confidence level with recurrence device network.Compared with prior art, accuracy of detection of the present invention is high, and testing result is accurately and reliably.
Description
Technical field
The present invention relates to a kind of phone housing defect inspection method, more particularly, to a kind of mobile phone based on deep learning outside
Shell defect inspection method.
Background technology
Popularization and its quick update, the phone housing product of industrial producing line with mobile phone, there is greatly product
Amount demand.It is normal on phone housing because of situations such as transport, production technology, accident from during the entire process of expecting final molding
Various defects (such as gouge, scuffing, scratch, heterochromatic unequal) are commonly present, and the product of these existing defects can influence its property
Or Consumer's Experience can be reduced, thus be not allow flow into market.Although in the past more than ten years in, industrial products production have
Greatly progressive and Production requirement increasingly increases, but the defects detection of related industries product is still relied on is accomplished manually, now main
The artificial visual of stream detects not only inefficiency, and examination criteria subjective factor is big, the automation that serious restriction industry is manufactured
Process, and artificial online defects detection can not only make cost increase, and also human resources are proposed with test.In recent years, it is based on
The automation defect inspection method of machine vision is of interest by numerous researchers, is also increasingly favored by manufacturer, but existing
Method accuracy of detection is low and time-consuming, it is impossible to meet real-time detection demand, and becoming the restriction machine substitution mankind carries out defect inspection
The principal element of survey.
Not yet have defects detection patent for phone housing at present, but exist for mobile phone liquid crystal screen curtain defects detection with
And to the defects detection of Mobile phone bottom plate connector.Existing mobile phone defects detection detection technique, mostly using traditional image at
Reason and identification technology, such as greyscale transformation, image binaryzation, rim detection, template matches etc.；And make use of more classical people
The operators such as work feature, such as SIFT, SURF, Haar, HOG, and image classification is carried out by neutral net or SVM classifier.
Defects detection wherein to mobile phone screen is that, by gathering liquid crystal display picture rich in detail, the image that will be collected carries out ash
Degree treatment, then enters ranks projection and row projection respectively to gray level image, according to the minimum of projection, will most start the figure for obtaining
As being divided into netted block of pixels image, then whole image is divided into multiple regions, each region includes multiple block of pixels, pin
Defective block of pixels is detected according to the gap of each block of pixels gray scale and the zone leveling gray scale to each region；So as to various pictures
The defects detection of element.
And the defects detection of Mobile phone bottom plate connector is then mainly carried out by template matches.First have to making standard
Mobile phone bottom plate connector gray level image template；Then, test image is treated to be pre-processed and greyscale transformation；And respectively to surveying
Attempting picture and template carries out SURF algorithm treatment and obtains characteristic point and affine change, with arest neighbors matching method matching characteristic point；It
Afterwards, error hiding characteristic point is eliminated with RANSAC models；Image to be tested is transformed into template yardstick sky further according to affine matrix
Between on, obtain a secondary new images, carry out binaryzation to the test image and standard picture after correction respectively, and by binary image
Subtracted each other；Finally, Morphological scale-space is carried out to error image, judges whether test image is qualified on this basis, if
It is defective, marking of defects position.
Similar is processed and defects detection mode of the feature extraction as Main Means with traditional images, has also applied to the sun
On the fields such as the defects detection of energy plate, the defects detection of rail, LED defects detections, but for the defects detection of phone housing
This problem, because phone housing defect is present, area is smaller, and extremely slightly, defective form is varied, with background contrasts
Not strong the characteristics of, above-mentioned traditional algorithm can not be well applied in the defects detection of phone housing, no matter from process time
Or on the precision of detection, can not all meet industrial demand.
The content of the invention
The purpose of the present invention is exactly to provide a kind of based on deep learning for the defect for overcoming above-mentioned prior art to exist
Phone housing defect inspection method.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of phone housing defect inspection method based on deep learning, the method comprises the following steps：
(1) obtain phone housing image to be detected and pre-processed；
(2) be input into pretreated image carries out defects detection and obtains mobile phone to the good defects detection model of training in advance
The position of existing defects on shell, and provide the confidence level that the position is defect；
Wherein, defects detection model is the depth network based on deep learning, including the feature extraction for cascading successively
Network and grader carry out feature extraction and obtain spy with device network, described feature extraction network is returned to the image for pre-processing
Levy image, described grader with return device network characteristic image is carried out classification recurrence obtain phone housing defective locations and
Confidence level.
Pretreatment specifically includes following steps in step (1)：
(101) phone housing image to be detected is carried out into size change over to being sized；
(102) image after size change in step (101) is carried out into rim detection and obtains edge image；
(103) Hough transformation is carried out to edge image and extracts detection zone and obtain strip image；
(104) strip image is entered line tilt correction and the strip image after correction is carried out into cutting and obtained with splicing
Square-shaped image.
Rim detection is carried out using Canny operators in step (102).
The training method of defects detection model is：
A () sets up described depth network；
B () gathers a large amount of phone housing images and carries out handmarking, iris out the region of existing defects, provides existing defects
The starting point coordinate and terminal point coordinate in region, and then obtain data sample；
C be input into data sample to depth network and carry out position and defect that feature extraction and classifying recurrence obtains defect by ()
Confidence level；
D defective locations and the confidence level of defect that () obtains step (c) are contrasted with the result of handmarking, so that
Each link weights in percentage regulation network, and then complete the training of depth network.
Described feature extraction network includes 5 feature extraction elementary cells for cascading successively, each feature extraction unit
Including the convolutional layer, local acknowledgement normalization layer, maximum pond layer and the average value pond layer that are sequentially connected；
Described convolutional layer is slided using convolution kernel on image, and convolution operation is carried out to image, so as to extract input figure
As feature obtains more rough characteristic pattern；
Described local acknowledgement's normalization layer using the field of the pixels of 3 pixel * 3 obtained in convolutional layer it is more rough
Slided on characteristic pattern, and the normalization of average and variance is carried out to the pixel value in each field, obtain not receiving illumination variation shadow
Loud rough characteristic pattern；
Described maximum pond layer using the pixels of 3 pixel * 3 field local acknowledgement normalization layer in obtain it is rough
Slided on characteristic pattern, and maximum is taken to all pixels value in each field, obtained with the more accurate of translation invariance
Characteristic pattern；
Described average pond layer using the field of the pixels of 3 pixel * 3 obtained in maximum pond layer it is more accurate
Slided on characteristic pattern, and all pixels value in each field is averaged, obtain having the accurate of robustness to miniature deformation
Characteristic pattern, described accurate characteristic pattern is the characteristic pattern of final corresponding feature extraction elementary cell output；
The feature extraction elementary cell final output characteristic image cascaded successively by 5.
Convolutional layer set-up mode in 5 feature extraction elementary cells for cascading successively is as follows：
In first feature extraction elementary cell, convolution kernel size is 7, for extracting larger feature, output characteristic figure
Number is 30；
In second feature extraction elementary cell, convolution kernel size is 5, and for extracting medium sized feature, output is special
It is 60 to levy map number；
In 3rd feature extraction elementary cell, convolution kernel size is 3, for extracting less feature, output characteristic figure
Number is 90；
In 4th feature extraction elementary cell, convolution kernel size is 3, for extracting minutia, output characteristic figure number
Mesh is 128；
In 5th feature extraction elementary cell, convolution kernel size is 3, for extracting minutia, output characteristic figure number
Mesh is 256.
The first full articulamentum and the second full articulamentum that described grader is cascaded successively with recurrence device network, described first
Full articulamentum input feature vector image, the second full articulamentum output end is connected with grader and returns device；
The first described full articulamentum is weighted to the characteristic image that feature extraction network is exported, obtain feature to
Amount；
The second described full articulamentum is weighted to the characteristic vector of the first full articulamentum output, refined and
The characteristic vector that feature is protruded；
Described grader judges the characteristic vector that the refinement of the second full articulamentum output and feature are protruded, judges
Whether belong to defect and provide the confidence level for belonging to defect；
Described recurrence device carries out recurrence treatment to the characteristic vector that the refinement of the second full articulamentum output and feature are protruded,
The positional information of the defect for being detected.
Compared with prior art, the invention has the advantages that：
(1) present invention sets up the depth network based on deep learning and is trained and obtains defects detection model, by this
Defects detection model carries out defects detection, and accuracy of detection is high, and testing result is accurately and reliably；
(2) present invention carries out a series of pretreatment to phone housing image to be detected, on the one hand reduces down-stream
Treating capacity, improves detection speed, has on the other hand carried out partial enlargement to the region for extracting, and makes to there may be the position of defect
More protrude, i.e., defect characteristic is more prominent, be ready characteristic extraction step below, it is final to improve recall rate reduction mistake
The rate of mistake；
(3) present invention carries out rim detection using Canny operators, and the effect of Canny edge extractings is good, suitable in setting
After threshold value, can make region to be detected with background area boundary substantially, improve the effect of subsequent defective detection；
(4) feature of present invention extracts network and uses 5 feature extraction elementary cells for cascading successively, and image is extracted respectively
In larger, medium, more detail feature, on the one hand avoid less convolutional layer to image characteristics extraction not enough fully and character modules
The shortcoming of paste, on the other hand simply use the feature that 5 convolutional layers are enough to extract phone housing defect, it is to avoid excessive convolutional layer
The problem of the huge amount of calculation brought, reduces occupancy resource and also saves overall recognition time.
Brief description of the drawings
Fig. 1 is the FB(flow block) of phone housing defect inspection method of the present invention based on deep learning；
Fig. 2 is the flow chart of phone housing image preprocessing to be detected of the invention；
Fig. 3 is the flow chart that Hough transformation of the present invention determines ROI；
Fig. 4 is the structural representation that feature of present invention extracts network；
Fig. 5 is grader of the present invention and the structural representation for returning device network；
Fig. 6 is the pretreated phone housing image to be detected of the present embodiment；
Fig. 7 obtains characteristic image for the present embodiment feature extraction；
Fig. 8 is the image that the present embodiment defects detection terminates output.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment
As shown in figure 1, a kind of phone housing defect inspection method based on deep learning, the method comprises the following steps：
(1) obtain phone housing image to be detected and pre-processed；
(2) be input into pretreated image carries out defects detection and obtains mobile phone to the good defects detection model of training in advance
The position of existing defects on shell, and provide the confidence level that the position is defect；
Wherein, defects detection model is the depth network based on deep learning, including the feature extraction for cascading successively
Network and grader carry out feature extraction and obtain characteristic pattern with device network, feature extraction network is returned to the image for pre-processing
Picture, grader carries out classification recurrence to characteristic image and obtains phone housing defective locations and confidence level with recurrence device network.
Pretreatment specifically includes following steps in step (1)：
(101) phone housing image to be detected is carried out into size change over to being sized；
(102) image after size change in step (101) is carried out into rim detection and obtains edge image；
(103) Hough transformation is carried out to edge image and extracts detection zone and obtain strip image；
(104) strip image is entered line tilt correction and the strip image after correction is carried out into cutting and obtained with splicing
Square-shaped image.
Rim detection is carried out using Canny operators in step (102).
The training method of defects detection model is：
A () sets up depth network；
B () gathers a large amount of phone housing images and carries out handmarking, iris out the region of existing defects, provides existing defects
The starting point coordinate and terminal point coordinate in region, and then obtain data sample；
C be input into data sample to depth network and carry out position and defect that feature extraction and classifying recurrence obtains defect by ()
Confidence level；
D defective locations and the confidence level of defect that () obtains step (c) are contrasted with the result of handmarking, so that
Each link weights in percentage regulation network, and then complete the training of depth network.
Feature extraction network include 5 feature extraction elementary cells for cascading successively, each feature extraction unit include according to
The convolutional layer of secondary connection, local acknowledgement normalize layer, maximum pond layer and average value pond layer；
Convolutional layer is slided using convolution kernel on image, and convolution operation is carried out to image, so as to extract input picture feature
Obtain more rough characteristic pattern；
The more rough characteristic pattern that local acknowledgement's normalization layer is obtained using the field of the pixels of 3 pixel * 3 in convolutional layer
Upper slip, and the normalization of average and variance is carried out to the pixel value in each field, obtain not by illumination variation influenceed it is thick
Characteristic pattern slightly；
The rough characteristic pattern that maximum pond layer is obtained using the field of the pixels of 3 pixel * 3 in local acknowledgement's normalization layer
Upper slip, and maximum is taken to all pixels value in each field, obtain the more accurate feature with translation invariance
Figure；
The more accurate characteristic pattern that average pond layer is obtained using the field of the pixels of 3 pixel * 3 in maximum pond layer
Upper slip, and all pixels value in each field is averaged, obtain the accurate feature for having robustness to miniature deformation
Figure, accurate characteristic pattern is the characteristic pattern of final corresponding feature extraction elementary cell output；
The feature extraction elementary cell final output characteristic image cascaded successively by 5.
Convolutional layer set-up mode in 5 feature extraction elementary cells for cascading successively is as follows：
In first feature extraction elementary cell, convolution kernel size is 7, for extracting larger feature, output characteristic figure
Number is 30；
In second feature extraction elementary cell, convolution kernel size is 5, and for extracting medium sized feature, output is special
It is 60 to levy map number；
In 3rd feature extraction elementary cell, convolution kernel size is 3, for extracting less feature, output characteristic figure
Number is 90；
In 4th feature extraction elementary cell, convolution kernel size is 3, for extracting minutia, output characteristic figure number
Mesh is 128；
In 5th feature extraction elementary cell, convolution kernel size is 3, for extracting minutia, output characteristic figure number
Mesh is 256.
The first full articulamentum and the second full articulamentum that grader is cascaded successively with recurrence device network, the described first full connection
Layer input feature vector image, the second full articulamentum output end is connected with grader and returns device；
First full articulamentum is weighted to the characteristic image that feature extraction network is exported, and obtains characteristic vector；
Second full articulamentum is weighted to the characteristic vector of the first full articulamentum output, is refined and feature is prominent
The characteristic vector for going out；
Grader judges the characteristic vector that the refinement of the second full articulamentum output and feature are protruded, judges whether to belong to
In defect and provide the confidence level for belonging to defect；
Return device carries out recurrence treatment to the characteristic vector that the refinement of the second full articulamentum output and feature are protruded, and is examined
The positional information of the defect measured.
According to above-mentioned narration, detailed technology explanation of the present invention is classified into following four part：
1. the extraction in region to be identified；
2. depth network is built；
3. sample labeling and model training；
4. defect recognition is carried out using existing model.
Specifically：
1st, the extraction in region to be identified：
The purpose of the step, due to that can there is the part without detection, such as Mobile phone casing clamper, and background etc. in image,
And there is the changes such as rotation, scaling in the image for collecting, then need to design a kind of algorithm to protrude housing region, on the one hand subtract
The treating capacity of few down-stream, improves detection speed, has on the other hand carried out partial enlargement to the region for extracting, and makes to deposit
More protruded in the position of defect, i.e., defect characteristic is more prominent, be ready characteristic extraction step below, final raising
Recall rate reduction error rate.
The algorithm of design and realization：This step is using the source images detection zone based on OpenCV (computer vision of increasing income storehouse)
Domain extracting method, the part without detection and the image rotation after cutting out are cropped by the original image collected from industrial camera
Turn, zoom to same size.The core of its algorithm is the image for collecting early stage industrial camera, by rim detection, suddenly
Husband's conversion etc. carries out coarse localization to interest region, and Slant Rectify is carried out on this basis, excludes the error that picture tilting band is come,
The degree of accuracy of interest zone location is improved, finally, the elongated interest region that will be extracted is split, be the defect inspection of next step
Survey is laid a solid foundation.Shown in testing process following 2, the original image that the input of the algorithm is collected for industrial camera, due to industry
The image resolution ratio of camera collection is high (about 1920*1440), in order to reduce the process time of algorithm, while also reducing noise
Influence, carries out 1/4 scaling to image first.Then because detection object is phone housing, its profile has straight line portion more long,
So by following rim detection and Hough transformation come detection of straight lines, and the straight line detected by judgement cross correlation come
Determine phone housing region, final output comprises only mobile phone shell to be detected and the unified picture of size, and specific algorithm details is such as
Under：
(1) rim detection
Image border is one of image key character information, and discontinuity is shown as in gray scale, is believed for analysis image
Breath and feature are significant, and this programme carries out rim detection using Canny operators to image.
Canny edge detection algorithms are a kind of algorithms that optimization thought is applied to image procossing, are calculated with conventional differential
Son is compared, and is not only possessed noise output higher and is compared, also quite reliable precision, and the effect of Canny edge extractings can be very
The good requirement for meeting this project, after suitable threshold value is set, can make regions to be detected bright with background area boundary
It is aobvious.Input is original image in this step, is output as edge image.
(2) Hough transformation is selected with ROI frames
The step is carried out in the result of rim detection, it is therefore an objective to which detection of straight lines is (because the edge of phone housing is big
Part is straight line) four steps are divided into, idiographic flow is as shown in Figure 3：
Step1：Edge extracting image to being exported in previous step carries out Hough transformation, can obtain a series of straight line,
Hough transformation is output as the polar diameter and polar angle of every straight line, and then can determine expression of the every straight line under x-y coordinate system
Formula:；
Step2：Determine intercept of this straight line on the image upper bound and image lower bound using the expression formula of every straight line,
It should be specifically noted that straight slope is infinitely great situation during this, in this case, this straight line is directly made in image
Intercept on bound is equal to polar diameter；
Step3：Determine four angle points of quadrangle ROI：The maximum of difference cut-off line intercept on image up-and-down boundary
And minimum value, four angle points of quadrangle ROI region can be obtained；
Step4:Four straight lines are constructed by four angle points, required for the closing quadrangle that four straight lines determine is us
ROI region, wherein ROI be area-of-interest.Input is in this step：Original image and edge image, are output as：Comprise only
The image in mobile phone shell region.
(3) picture correction is inclined
In real process, the phone housing image that camera is collected is controlled to there is rotation, translation and chi by mechanical arm
The problems such as degree change, for convenience detection of subsequent step in this programme, using acquired angular coordinate, is schemed to inclining
Piece is corrected, and finally for the picture of different inputs, makes output picture for a size yardstick is identical, without spin, perpendicular to
Horizontal image.Input is in this step：Original image and edge image, are output as：Comprise only the image in mobile phone shell region.
Mobile phone shell regional correction algorithm utilizes the upper left corner, the upper right corner and the lower left corner in region in (2) output in this programme
Three angle points, the transformation matrix M required for obtaining affine transformation, the thinking of mapping is that two angle points are constant above, the horizontal seat of angle point
Mark is mapped to identical with the abscissa in the upper left corner.Setting tolerance threshold value, is carried out affine to the inclination picture more than tolerance
Conversion.Input is in this step：Comprise only the image (now there is rotation and dimensional variation) in mobile phone shell region, output：Containing only
There is the image (unified yardstick, without spin) of phone area.
(4) phone housing image segmentation：
By the correction of above-mentioned image preprocessing, Hough transformation and tilted image, can reasonably accurately extract
Phone housing region, but because the region is elongated rectangle, and the deep learning network of this programme design aligns square chart
The Detection results of picture are preferable, so as to extract before result inputs to deep learning defect detecting system, it is necessary to be carried out to it
Dividing processing, and then the detection efficiency of defect detecting system is improved, reduce the pressure of defect detecting system.Input is in this step
(3) image (rectangle) for comprising only mobile phone shell region of generation in, output：Mobile phone shell is comprised only by certain regular well cutting
The image (square) in region.
2nd, depth network is built：
The purpose of the step is that the region to be detected extracted to previous step carries out feature extraction, obtains altimetric image to be checked
Validity feature, classified so as to subsequent network, returned.(because (being also neutral net for a deep learning network
It is a kind of), be characterized in the foundation of classification, Feature Selection it is more representative, then classification results are better).
Advantage：Compared to traditional method based on the artificial feature extraction such as SIFT, HOG, this programme uses convolutional network
Carry out Automatic Feature Extraction.The artificial feature extracted tends not to preferably completely represent out by the feature of a pictures, and
Automatic Feature Extraction network then can efficiently extract the key character of image.
This programme uses forefront algorithm --- the Faster-RCNN algorithm ideas in deep learning object detection field,
With reference to image procossing, build and be suitable to the deep learning network of defects detection.Important Thought is basic by four of target detection
Step (candidate region generates, feature extraction, classification, position refine) is unified to be arrived within same depth network frame.All meters
Calculate without repeating, completed completely in GPU, so as to substantially increase the speed of service.
In order to generate candidate frame to be detected and classification for after provides foundation with returning, it is necessary first to piece image
Carry out feature extraction.Its structure is as shown in Figure 4.Convolutional layer：Slided on image using convolution kernel, convolution behaviour is carried out to image
Make.Convolution operation can be used to do image rim detection, sharpen, obscure etc., the extraction to characteristics of image is can be used as herein.
In this programme, each layer is controlled to extract different features by controlling the size of convolution kernel, and each convolution kernel (is exactly individual square
Battle array) in each element by train part determine (i.e. train when need modification weight, in feature extraction network, only roll up
The convolution kernel of lamination is to need training).
Convolution operation is carried out to image, be to each pixel in image-region respectively at convolution kernel (weight matrix) each
Element correspondence be multiplied, all sum of products as regional center pixel new value.
ReLU (linearity rectification function) activation primitive：Convergence rate of the network in training can be accelerated (by neurosurgeon
It is considered to more conform to a kind of function of neuron signal exiting principle), and because computing is simple, calculation resources can be saved.
This programme uses Leaky ReLU activation primitives.
Local acknowledgement normalizes layer：Each pixel value to the region of 3*3 (this programme using) in image domains is carried out
The normalization of value and variance, reaches elimination background and illumination effect, the effect of prominent features.The normalization of average and variance i.e.,
Each pixel value collection in original image field is normalized to the data that average is 0, variance 1, such data have good
Measurement property.Normalizing formula is：
Z=χ-μ/σ
Wherein z represents the value after normalization, and x represents the pixel value before normalization, and μ represents pixel in input picture field
The average of value, σ represents the variance of pixel value in input picture field.
Maximum pond layer：I.e. to (this programme uses the region of 3*3) all pixels value maximizing in neighborhood, can reduce
Data volume, while realizing the translation invariance to the feature extracted.
Average pond layer：I.e. to (this programme uses the region of 3*3) all pixels value ball average value in neighborhood, can reduce
Data volume, while strengthening the robustness of the feature to miniature deformation of extraction.
By such network, being often input into a pictures just can obtain 256 characteristic patterns (exporting), what is extracted
On characteristic pattern, candidate frame is generated, the effect of candidate frame is for judging whether a certain region there may be defect, then to this
A little candidate frames are classified and are returned, and whether the effect of classification is to judge include object to be detected in each candidate frame；Return
Effect be that will be possible to the candidate frame comprising examined object to be combined, merge, and position is modified.Waited
Select frame to classify to be completed by two full articulamentums with the network structure for returning, the nodes of each full articulamentum are 1024, and it is opened up
Flutter structure as shown in Figure 5.In such as network of Fig. 5, layer (cls_score) final output of classifying belongs to prospect on each position
The probability of (examined object) and background；Window to be returned and should translate contracting on each position of layer (bbox_pred) final output
The parameter put.
For each position, classification layer exports the probability for belonging to foreground and background from 256 dimensional features；Window is returned
Return layer that 4 translation zooming parameters are exported from 256 dimensional features.
Thus, the feature extraction to a pictures is just completed, while candidate frame is generated, and it is to be detected to that may include
The candidate frame of object (i.e. defect) is classified, and gives regression parameter.Full articulamentum：I.e. any one of this layer is saved
Point, all nodes with next layer have connection, and so next layer of output is just relevant (i.e. common with all inputs of last layer
BP neural network)；And because first full articulamentum is connected with the best middle output of feature extraction network, that is, cause final
Output it is relevant with all features for extracting；The weight of each node of full articulamentum needs by training determination that (method is
Back propagation).
ReLU layers：ReLu layers in parameter and effect same " feature extraction network ", in the portion, as full articulamentum
Activation primitive.
Dropout layers：Due to the presence of full articulamentum so that final output is related to each feature, but trains sometimes
Sample may not represent identified all situations, it is possible to which (over-fitting is exactly can to training sample over-fitting situation occurs
To there is an accuracy very high, and in practical application, situation to not seen in training sample, can not be imitated well
Really).Dropout layers prevents over-fitting by randomly hiding the node increase of some full articulamentums, strengthens generalization ability.
SoftmaxWithLoss layers：The layer only exists in training, is the evaluation to full articulamentum final output result,
For calculating the gap between grader classification results and actual value.
In such as network of Fig. 5, prospect (examined object) and background are belonged on each position of grader final output
Probability；The parameter of scaling should be translated on each position of window device final output.
For each position, classification layer exports the probability for belonging to foreground and background from 256 dimensional features；Window is returned
Return layer that 4 translation zooming parameters are exported from 256 dimensional features.
Thus, the feature extraction to a pictures is just completed, while candidate frame is generated, and it is to be detected to that may include
The candidate frame of object (i.e. defect) is classified, and gives regression parameter.
3rd, sample labeling and model training：
First to camera acquisition to image carry out manual identified, iris out the region of existing defects, provide existing defects
The starting point coordinate and terminal point coordinate in region.To then enter in the depth network described in a large amount of sections of sample feeding 3.2 for having marked
Row training.The training process is general to be completed on GPU.Its training process can be summarized as：
A every image in training set) is primarily looked at：
A. to the true value candidate region that each is demarcated, overlap the maximum candidate frame of ratio and be designated as prospect sample；
B. to a) remaining candidate frame, if it is more than 0.7 with certain demarcation overlap proportion, it is designated as prospect sample；If
Its overlap proportion demarcated with any one is both less than 0.3, is designated as background sample；
C. to a), b) remaining candidate frame, is discarded；
D. the candidate frame across image boundary is discarded.
B) then by the classification described in all features extracted and the candidate frame of generation feeding 3.2 and Recurrent networks
In, just can obtain each candidate frame and belong to the probability of foreground and background and be used to the parameter for returning.The knot that these are obtained
Fruit contrasted with actual value (result of handmarking), evaluated using two cost functions described in 3.2 discre value with
The difference of actual value, finally by the inverse iteration algorithm that neutral net is general, each link weights in adjustment network so that
Error in classification and regression error described in 3.2 are minimized simultaneously.
4th, defect recognition is carried out using existing model：
In this programme, can directly input original image carries out defects detection, whether there is in the final output image
Defect, if existing defects, provides the coordinate and its confidence level of defective locations.
The present embodiment arbitrarily one picture to be detected of input, the picture is the side of mobile phone shell, carries out pre-processing first
To pretreated phone housing image to be detected, obtained such as figure as shown in fig. 6, being then fed into depth network and carrying out feature extraction
Characteristic image shown in 7, finally obtains the image that the defects detection shown in Fig. 8 terminates output, exists at mark in the image and lacks
Sunken coordinate position and the confidence level for belonging to defect.
Claims (7)
1. a kind of phone housing defect inspection method based on deep learning, it is characterised in that the method comprises the following steps：
(1) obtain phone housing image to be detected and pre-processed；
(2) be input into pretreated image carries out defects detection and obtains phone housing to the good defects detection model of training in advance
The position of upper existing defects, and provide the confidence level that the position is defect；
Wherein, defects detection model is the depth network based on deep learning, including the feature extraction network for cascading successively
And grader carries out feature extraction and obtains characteristic pattern with device network, described feature extraction network is returned to the image for pre-processing
Picture, described grader carries out classification recurrence to characteristic image and obtains phone housing defective locations and confidence with recurrence device network
Degree.
2. a kind of phone housing defect inspection method based on deep learning according to claim 1, it is characterised in that step
Suddenly pretreatment specifically includes following steps in (1)：
(101) phone housing image to be detected is carried out into size change over to being sized；
(102) image after size change in step (101) is carried out into rim detection and obtains edge image；
(103) Hough transformation is carried out to edge image and extracts detection zone and obtain strip image；
(104) strip image is entered line tilt correction and the strip image after correction is carried out into cutting and obtain pros with splicing
Shape image.
3. a kind of phone housing defect inspection method based on deep learning according to claim 2, it is characterised in that step
Suddenly rim detection is carried out using Canny operators in (102).
4. a kind of phone housing defect inspection method based on deep learning according to claim 1, it is characterised in that lack
Fall into detection model training method be：
A () sets up described depth network；
B () gathers a large amount of phone housing images and carries out handmarking, iris out the region of existing defects, provides existing defects region
Starting point coordinate and terminal point coordinate, and then obtain data sample；
(c) by data sample be input into depth network carry out feature extraction and classifying return obtain the position of defect and putting for defect
Reliability；
D defective locations and the confidence level of defect that () obtains step (c) are contrasted with the result of handmarking, so as to adjust
Each link weights in depth network, and then complete the training of depth network.
5. a kind of phone housing defect inspection method based on deep learning according to claim 1, it is characterised in that institute
The feature extraction network stated includes 5 feature extraction elementary cells for cascading successively, and each feature extraction unit includes connecting successively
The convolutional layer that connects, local acknowledgement normalization layer, maximum pond layer and average value pond layer；
Described convolutional layer is slided using convolution kernel on image, and convolution operation is carried out to image, so that it is special to extract input picture
Obtain more rough characteristic pattern；
The more rough feature that described local acknowledgement's normalization layer is obtained using the field of the pixels of 3 pixel * 3 in convolutional layer
Slided on figure, and the normalization of average and variance is carried out to the pixel value in each field, obtain what is do not influenceed by illumination variation
Rough characteristic pattern；
The rough feature that described maximum pond layer is obtained using the field of the pixels of 3 pixel * 3 in local acknowledgement's normalization layer
Slided on figure, and maximum is taken to all pixels value in each field, obtain more accurate special with translation invariance
Levy figure；
The more accurate feature that described average pond layer is obtained using the field of the pixels of 3 pixel * 3 in maximum pond layer
Slided on figure, and all pixels value in each field is averaged, obtain the accurate spy for having robustness to miniature deformation
Figure is levied, described accurate characteristic pattern is the characteristic pattern of final corresponding feature extraction elementary cell output；
The feature extraction elementary cell final output characteristic image cascaded successively by 5.
6. a kind of phone housing defect inspection method based on deep learning according to claim 5, it is characterised in that 5
Convolutional layer set-up mode in the individual feature extraction elementary cell for cascading successively is as follows：
In first feature extraction elementary cell, convolution kernel size is 7, for extracting larger feature, output characteristic map number
It is 30；
In second feature extraction elementary cell, convolution kernel size is 5, for extracting medium sized feature, output characteristic figure
Number is 60；
In 3rd feature extraction elementary cell, convolution kernel size is 3, for extracting less feature, output characteristic map number
It is 90；
In 4th feature extraction elementary cell, convolution kernel size is 3, and for extracting minutia, output characteristic map number is
128；
In 5th feature extraction elementary cell, convolution kernel size is 3, and for extracting minutia, output characteristic map number is
256。
7. a kind of phone housing defect inspection method based on deep learning according to claim 1, it is characterised in that institute
The first full articulamentum and the second full articulamentum that the grader stated is cascaded successively with recurrence device network, the first full articulamentum are defeated
Enter characteristic image, the second full articulamentum output end is connected with grader and returns device；
The first described full articulamentum is weighted to the characteristic image that feature extraction network is exported, and obtains characteristic vector；
The second described full articulamentum is weighted to the characteristic vector of the first full articulamentum output, is refined and feature
Prominent characteristic vector；
Described grader judges the characteristic vector that the refinement of the second full articulamentum output and feature are protruded, judges whether
Belong to defect and provide the confidence level for belonging to defect；
Described recurrence device carries out recurrence treatment to the characteristic vector that the refinement of the second full articulamentum output and feature are protruded, and obtains
The positional information of the defect for detecting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710034677.0A CN106875381B (en) | 2017-01-17 | 2017-01-17 | Mobile phone shell defect detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710034677.0A CN106875381B (en) | 2017-01-17 | 2017-01-17 | Mobile phone shell defect detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106875381A true CN106875381A (en) | 2017-06-20 |
| CN106875381B CN106875381B (en) | 2020-04-28 |
Family
ID=59158719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710034677.0A Active CN106875381B (en) | 2017-01-17 | 2017-01-17 | Mobile phone shell defect detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106875381B (en) |
Cited By (62)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481231A (en) * | 2017-08-17 | 2017-12-15 | 广东工业大学 | A kind of handware defect classifying identification method based on depth convolutional neural networks |
| CN107730504A (en) * | 2017-10-17 | 2018-02-23 | 太原理工大学 | Image partition method based on improved BP |
| CN108009990A (en) * | 2017-11-30 | 2018-05-08 | 北京小米移动软件有限公司 | Tilt object processing method and device |
| CN108009581A (en) * | 2017-11-30 | 2018-05-08 | 中国地质大学（武汉） | A kind of method for crack based on CNN, equipment and storage device |
| CN108021914A (en) * | 2017-12-27 | 2018-05-11 | 清华大学 | Printed matter character zone extracting method based on convolutional neural networks |
| CN108229561A (en) * | 2018-01-03 | 2018-06-29 | 北京先见科技有限公司 | Particle product defect detection method based on deep learning |
| CN108362703A (en) * | 2017-12-14 | 2018-08-03 | 北京木业邦科技有限公司 | A kind of veneer detection method and detection device based on artificial intelligence |
| CN108362702A (en) * | 2017-12-14 | 2018-08-03 | 北京木业邦科技有限公司 | A kind of defect of veneer detection method, system and equipment based on artificial intelligence |
| CN108355981A (en) * | 2018-01-08 | 2018-08-03 | 西安交通大学 | A kind of battery connector quality determining method based on machine vision |
| CN108460413A (en) * | 2018-02-12 | 2018-08-28 | 北京市商汤科技开发有限公司 | Image processing method, the training method of neural network, device and equipment |
| CN108510469A (en) * | 2018-02-08 | 2018-09-07 | 郑州瑞富德机电科技有限公司 | A kind of bracket detection method and device |
| CN108596167A (en) * | 2018-04-19 | 2018-09-28 | 上海亿锎智能科技有限公司 | A kind of elongated optical character word system for rapidly identifying and method, model |
| CN108685158A (en) * | 2018-05-28 | 2018-10-23 | 杭州安脉盛智能技术有限公司 | Filament cutter intelligent maintenance method and system based on automatic optics inspection technology |
| CN109087294A (en) * | 2018-07-31 | 2018-12-25 | 深圳辰视智能科技有限公司 | A kind of product defects detection method, system and computer readable storage medium |
| CN109117858A (en) * | 2017-06-26 | 2019-01-01 | 北京金风科创风电设备有限公司 | Method and device for monitoring icing of blades of wind driven generator |
| CN109146873A (en) * | 2018-09-04 | 2019-01-04 | 凌云光技术集团有限责任公司 | A kind of display screen defect intelligent detecting method and device based on study |
| CN109242830A (en) * | 2018-08-18 | 2019-01-18 | 苏州翔升人工智能科技有限公司 | A kind of machine vision technique detection method based on deep learning |
| CN109239075A (en) * | 2018-08-27 | 2019-01-18 | 北京百度网讯科技有限公司 | Battery detection method and device |
| CN109544501A (en) * | 2018-03-22 | 2019-03-29 | 广东电网有限责任公司清远供电局 | A kind of transmission facility defect inspection method based on unmanned plane multi-source image characteristic matching |
| CN109636772A (en) * | 2018-10-25 | 2019-04-16 | 同济大学 | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning |
| CN109670518A (en) * | 2018-12-25 | 2019-04-23 | 浙江大学常州工业技术研究院 | A kind of method on the boundary of object in measurement picture |
| CN109711474A (en) * | 2018-12-24 | 2019-05-03 | 中山大学 | A kind of aluminium material surface defects detection algorithm based on deep learning |
| CN109726754A (en) * | 2018-12-25 | 2019-05-07 | 浙江大学昆山创新中心 | A kind of LCD screen defect identification method and device |
| CN109859169A (en) * | 2019-01-03 | 2019-06-07 | 重庆信息通信研究院 | A kind of TFT-LCD mould group white seal detection method based on convolutional neural networks |
| CN109919941A (en) * | 2019-03-29 | 2019-06-21 | 深圳市奥特立德自动化技术有限公司 | Internal screw thread defect inspection method, device, system, equipment and medium |
| CN109934811A (en) * | 2019-03-08 | 2019-06-25 | 中国科学院光电技术研究所 | A kind of optical element surface defect inspection method based on deep learning |
| CN110021005A (en) * | 2018-01-05 | 2019-07-16 | 财团法人工业技术研究院 | Circuit board flaw screening technique and its device and computer-readable recording medium |
| CN110119677A (en) * | 2019-03-28 | 2019-08-13 | 东南大学 | Carbon fiber composite core cable damage testing method based on image classification network |
| CN110648323A (en) * | 2019-09-26 | 2020-01-03 | 上海御微半导体技术有限公司 | Defect detection classification system and method thereof |
| CN110675399A (en) * | 2019-10-28 | 2020-01-10 | 上海悦易网络信息技术有限公司 | Screen appearance flaw detection method and equipment |
| CN110706205A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(重庆)科技有限公司 | Method for detecting cloth hole-breaking defect by using computer vision technology |
| CN110781888A (en) * | 2019-10-25 | 2020-02-11 | 北京字节跳动网络技术有限公司 | Method and device for regressing screen in video picture, readable medium and electronic equipment |
| CN110781931A (en) * | 2019-10-14 | 2020-02-11 | 国家广播电视总局广播电视科学研究院 | Ultrahigh-definition film source conversion curve detection method for local feature extraction and fusion |
| CN110796647A (en) * | 2019-10-28 | 2020-02-14 | 上海悦易网络信息技术有限公司 | Method and device for detecting defects of screen area of electronic device |
| CN110827249A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment backboard appearance flaw detection method and equipment |
| CN110827246A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment frame appearance flaw detection method and equipment |
| CN110910360A (en) * | 2019-11-14 | 2020-03-24 | 腾讯云计算（北京）有限责任公司 | Power grid image positioning method and image positioning model training method |
| CN110942450A (en) * | 2019-11-19 | 2020-03-31 | 武汉大学 | Multi-production-line real-time defect detection method based on deep learning |
| CN110969620A (en) * | 2019-12-19 | 2020-04-07 | 中国农业大学 | Method and device for detecting magnetic shoe ripple defects |
| CN110992361A (en) * | 2019-12-25 | 2020-04-10 | 创新奇智(成都)科技有限公司 | Engine fastener detection system and detection method based on cost balance |
| CN111062417A (en) * | 2019-11-19 | 2020-04-24 | 武汉大学 | Lie-Mean-based flat shell defect detection method and system |
| CN111079629A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon center plate out-of-position fault image identification method |
| WO2020137222A1 (en) * | 2018-12-28 | 2020-07-02 | オムロン株式会社 | Defect inspecting device, defect inspecting method, and program for same |
| CN111369545A (en) * | 2020-03-10 | 2020-07-03 | 创新奇智(重庆)科技有限公司 | Edge defect detection method, device, model, equipment and readable storage medium |
| CN111400040A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Industrial Internet system based on deep learning and edge calculation and working method |
| CN111768404A (en) * | 2020-07-08 | 2020-10-13 | 北京滴普科技有限公司 | Mask appearance defect detection system, method and device and storage medium |
| WO2020211823A1 (en) * | 2019-04-18 | 2020-10-22 | 杭州海康威视数字技术股份有限公司 | Model training method and device, and defect detection method and device |
| CN111833328A (en) * | 2020-07-14 | 2020-10-27 | 汪俊 | Aircraft engine blade surface defect detection method based on deep learning |
| CN112132196A (en) * | 2020-09-14 | 2020-12-25 | 中山大学 | Cigarette case defect identification method combining deep learning and image processing |
| CN112149594A (en) * | 2020-09-29 | 2020-12-29 | 同济大学 | Urban construction assessment method based on deep learning and high-resolution satellite images |
| WO2021057395A1 (en) * | 2019-09-29 | 2021-04-01 | 五邑大学 | Heel type identification method, device, and storage medium |
| CN112907519A (en) * | 2021-01-29 | 2021-06-04 | 广州信邦智能装备股份有限公司 | Metal curved surface defect analysis system and method based on deep learning |
| CN113112497A (en) * | 2021-05-06 | 2021-07-13 | 合肥中科迪宏自动化有限公司 | Industrial appearance defect detection method based on zero sample learning, electronic device and storage medium |
| CN113168687A (en) * | 2019-01-22 | 2021-07-23 | 株式会社日立高新技术 | Image evaluation apparatus and method |
| CN113192061A (en) * | 2021-05-25 | 2021-07-30 | 安徽睿微信息科技有限公司 | LED package appearance detection image extraction method and device, electronic equipment and storage medium |
| CN113240647A (en) * | 2021-05-18 | 2021-08-10 | 哈尔滨工业大学（深圳） | Mobile phone shell rear cover defect detection method and system based on deep learning |
| CN113302649A (en) * | 2018-10-16 | 2021-08-24 | 香港中文大学 | Method, device and system for automatic diagnosis |
| CN113610754A (en) * | 2021-06-28 | 2021-11-05 | 浙江文谷科技有限公司 | Defect detection method and system based on Transformer |
| TWI745767B (en) * | 2019-10-18 | 2021-11-11 | 汎思數據股份有限公司 | Optical inspection secondary image classification method |
| CN113705442A (en) * | 2021-10-09 | 2021-11-26 | 广东博媒广告传播有限公司 | Outdoor large-board advertising picture monitoring and identifying system and method |
| CN113744274A (en) * | 2021-11-08 | 2021-12-03 | 深圳市巨力方视觉技术有限公司 | Product appearance defect detection method and device and storage medium |
| CN115147409A (en) * | 2022-08-30 | 2022-10-04 | 深圳市欣冠精密技术有限公司 | Mobile phone shell production quality detection method based on machine vision |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102218406A (en) * | 2011-01-04 | 2011-10-19 | 华南理工大学 | Intelligent detection device of defects of mobile phone outer shell based on machine vision |
| CN105118044A (en) * | 2015-06-16 | 2015-12-02 | 华南理工大学 | Method for automatically detecting defects of wheel-shaped cast product |
-
2017
- 2017-01-17 CN CN201710034677.0A patent/CN106875381B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102218406A (en) * | 2011-01-04 | 2011-10-19 | 华南理工大学 | Intelligent detection device of defects of mobile phone outer shell based on machine vision |
| CN105118044A (en) * | 2015-06-16 | 2015-12-02 | 华南理工大学 | Method for automatically detecting defects of wheel-shaped cast product |
Non-Patent Citations (2)
| Title |
|---|
| D.SOUKUP ET AL.: "Convolutional Neural Networks for Steel Surface Defect Detection from Photometric Stereo Images", 《SPRINGER INTERNATIONAL PUBLISHING SWITZERLAND 2014》 * |
| XIU LI ET AL.: "Fast Accurate Fish Detection and Recognition of Underwater Images with Fast R-CNN", 《IEEE》 * |
Cited By (88)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109117858B (en) * | 2017-06-26 | 2024-02-13 | 北京金风科创风电设备有限公司 | Method and device for monitoring icing of wind driven generator blade |
| CN109117858A (en) * | 2017-06-26 | 2019-01-01 | 北京金风科创风电设备有限公司 | Method and device for monitoring icing of blades of wind driven generator |
| WO2019000818A1 (en) * | 2017-06-26 | 2019-01-03 | 北京金风科创风电设备有限公司 | Method and device for monitoring icing of wind turbine blade |
| EP3457326A4 (en) * | 2017-06-26 | 2019-09-25 | Beijing Goldwind Science & Creation Windpower Equipment Co., Ltd. | Method and device for monitoring icing of wind turbine blade |
| US11286913B2 (en) * | 2017-06-26 | 2022-03-29 | Beijing Goldwind Science & Creation Windpower Equipment Co., Ltd. | Method and apparatus for monitoring formation of ice on wind turbine blade |
| CN107481231A (en) * | 2017-08-17 | 2017-12-15 | 广东工业大学 | A kind of handware defect classifying identification method based on depth convolutional neural networks |
| CN107730504A (en) * | 2017-10-17 | 2018-02-23 | 太原理工大学 | Image partition method based on improved BP |
| CN108009581A (en) * | 2017-11-30 | 2018-05-08 | 中国地质大学（武汉） | A kind of method for crack based on CNN, equipment and storage device |
| CN108009990A (en) * | 2017-11-30 | 2018-05-08 | 北京小米移动软件有限公司 | Tilt object processing method and device |
| CN108362702A (en) * | 2017-12-14 | 2018-08-03 | 北京木业邦科技有限公司 | A kind of defect of veneer detection method, system and equipment based on artificial intelligence |
| CN108362703A (en) * | 2017-12-14 | 2018-08-03 | 北京木业邦科技有限公司 | A kind of veneer detection method and detection device based on artificial intelligence |
| CN108021914A (en) * | 2017-12-27 | 2018-05-11 | 清华大学 | Printed matter character zone extracting method based on convolutional neural networks |
| CN108021914B (en) * | 2017-12-27 | 2020-07-28 | 清华大学 | Method for extracting character area of printed matter based on convolutional neural network |
| CN108229561A (en) * | 2018-01-03 | 2018-06-29 | 北京先见科技有限公司 | Particle product defect detection method based on deep learning |
| CN110021005B (en) * | 2018-01-05 | 2022-03-15 | 财团法人工业技术研究院 | Method and device for screening defects of circuit board and computer readable recording medium |
| CN110021005A (en) * | 2018-01-05 | 2019-07-16 | 财团法人工业技术研究院 | Circuit board flaw screening technique and its device and computer-readable recording medium |
| CN108355981A (en) * | 2018-01-08 | 2018-08-03 | 西安交通大学 | A kind of battery connector quality determining method based on machine vision |
| CN108510469A (en) * | 2018-02-08 | 2018-09-07 | 郑州瑞富德机电科技有限公司 | A kind of bracket detection method and device |
| CN108460413A (en) * | 2018-02-12 | 2018-08-28 | 北京市商汤科技开发有限公司 | Image processing method, the training method of neural network, device and equipment |
| CN109544501A (en) * | 2018-03-22 | 2019-03-29 | 广东电网有限责任公司清远供电局 | A kind of transmission facility defect inspection method based on unmanned plane multi-source image characteristic matching |
| CN108596167B (en) * | 2018-04-19 | 2020-08-21 | 上海亿锎智能科技有限公司 | Variable-length optical character fast recognition system, method and device |
| CN108596167A (en) * | 2018-04-19 | 2018-09-28 | 上海亿锎智能科技有限公司 | A kind of elongated optical character word system for rapidly identifying and method, model |
| CN108685158A (en) * | 2018-05-28 | 2018-10-23 | 杭州安脉盛智能技术有限公司 | Filament cutter intelligent maintenance method and system based on automatic optics inspection technology |
| CN108685158B (en) * | 2018-05-28 | 2021-11-09 | 杭州安脉盛智能技术有限公司 | Intelligent filament cutter maintenance method and system based on automatic optical detection technology |
| CN109087294A (en) * | 2018-07-31 | 2018-12-25 | 深圳辰视智能科技有限公司 | A kind of product defects detection method, system and computer readable storage medium |
| CN109242830A (en) * | 2018-08-18 | 2019-01-18 | 苏州翔升人工智能科技有限公司 | A kind of machine vision technique detection method based on deep learning |
| CN109239075A (en) * | 2018-08-27 | 2019-01-18 | 北京百度网讯科技有限公司 | Battery detection method and device |
| CN109146873B (en) * | 2018-09-04 | 2020-12-29 | 凌云光技术股份有限公司 | Learning-based intelligent detection method and device for defects of display screen |
| CN109146873A (en) * | 2018-09-04 | 2019-01-04 | 凌云光技术集团有限责任公司 | A kind of display screen defect intelligent detecting method and device based on study |
| CN113302649A (en) * | 2018-10-16 | 2021-08-24 | 香港中文大学 | Method, device and system for automatic diagnosis |
| CN109636772A (en) * | 2018-10-25 | 2019-04-16 | 同济大学 | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning |
| CN109711474A (en) * | 2018-12-24 | 2019-05-03 | 中山大学 | A kind of aluminium material surface defects detection algorithm based on deep learning |
| CN109711474B (en) * | 2018-12-24 | 2023-01-17 | 中山大学 | Aluminum product surface defect detection algorithm based on deep learning |
| CN109670518A (en) * | 2018-12-25 | 2019-04-23 | 浙江大学常州工业技术研究院 | A kind of method on the boundary of object in measurement picture |
| CN109726754A (en) * | 2018-12-25 | 2019-05-07 | 浙江大学昆山创新中心 | A kind of LCD screen defect identification method and device |
| CN109670518B (en) * | 2018-12-25 | 2022-09-23 | 浙江大学常州工业技术研究院 | Method for measuring boundary of target object in picture |
| WO2020137222A1 (en) * | 2018-12-28 | 2020-07-02 | オムロン株式会社 | Defect inspecting device, defect inspecting method, and program for same |
| US11830174B2 (en) | 2018-12-28 | 2023-11-28 | Omron Corporation | Defect inspecting device, defect inspecting method, and storage medium |
| JP2020106467A (en) * | 2018-12-28 | 2020-07-09 | オムロン株式会社 | Defect inspection device, defect inspection method, and program therefor |
| CN109859169B (en) * | 2019-01-03 | 2022-12-16 | 重庆信息通信研究院 | TFT-LCD module white print detection method based on convolutional neural network |
| CN109859169A (en) * | 2019-01-03 | 2019-06-07 | 重庆信息通信研究院 | A kind of TFT-LCD mould group white seal detection method based on convolutional neural networks |
| CN113168687A (en) * | 2019-01-22 | 2021-07-23 | 株式会社日立高新技术 | Image evaluation apparatus and method |
| CN109934811B (en) * | 2019-03-08 | 2022-12-30 | 中国科学院光电技术研究所 | Optical element surface defect detection method based on deep learning |
| CN109934811A (en) * | 2019-03-08 | 2019-06-25 | 中国科学院光电技术研究所 | A kind of optical element surface defect inspection method based on deep learning |
| CN110119677A (en) * | 2019-03-28 | 2019-08-13 | 东南大学 | Carbon fiber composite core cable damage testing method based on image classification network |
| CN110119677B (en) * | 2019-03-28 | 2022-03-11 | 东南大学 | Image classification network-based carbon fiber composite core cable damage detection method |
| CN109919941A (en) * | 2019-03-29 | 2019-06-21 | 深圳市奥特立德自动化技术有限公司 | Internal screw thread defect inspection method, device, system, equipment and medium |
| WO2020211823A1 (en) * | 2019-04-18 | 2020-10-22 | 杭州海康威视数字技术股份有限公司 | Model training method and device, and defect detection method and device |
| CN110706205A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(重庆)科技有限公司 | Method for detecting cloth hole-breaking defect by using computer vision technology |
| CN110648323A (en) * | 2019-09-26 | 2020-01-03 | 上海御微半导体技术有限公司 | Defect detection classification system and method thereof |
| WO2021057395A1 (en) * | 2019-09-29 | 2021-04-01 | 五邑大学 | Heel type identification method, device, and storage medium |
| CN110781931A (en) * | 2019-10-14 | 2020-02-11 | 国家广播电视总局广播电视科学研究院 | Ultrahigh-definition film source conversion curve detection method for local feature extraction and fusion |
| CN110781931B (en) * | 2019-10-14 | 2022-03-08 | 国家广播电视总局广播电视科学研究院 | Ultrahigh-definition film source conversion curve detection method for local feature extraction and fusion |
| TWI745767B (en) * | 2019-10-18 | 2021-11-11 | 汎思數據股份有限公司 | Optical inspection secondary image classification method |
| CN110781888A (en) * | 2019-10-25 | 2020-02-11 | 北京字节跳动网络技术有限公司 | Method and device for regressing screen in video picture, readable medium and electronic equipment |
| JP2022539909A (en) * | 2019-10-28 | 2022-09-13 | 上海万物新生▲環▼保科技集団有限公司 | Electronic device frame appearance defect inspection method and apparatus |
| CN110675399A (en) * | 2019-10-28 | 2020-01-10 | 上海悦易网络信息技术有限公司 | Screen appearance flaw detection method and equipment |
| WO2021082919A1 (en) * | 2019-10-28 | 2021-05-06 | 上海悦易网络信息技术有限公司 | Defect detecting method and equipment for screen region of electronic equipment |
| CN110796647A (en) * | 2019-10-28 | 2020-02-14 | 上海悦易网络信息技术有限公司 | Method and device for detecting defects of screen area of electronic device |
| CN110827249A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment backboard appearance flaw detection method and equipment |
| CN110827246A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment frame appearance flaw detection method and equipment |
| CN110910360B (en) * | 2019-11-14 | 2023-06-13 | 腾讯云计算（北京）有限责任公司 | Positioning method of power grid image and training method of image positioning model |
| CN110910360A (en) * | 2019-11-14 | 2020-03-24 | 腾讯云计算（北京）有限责任公司 | Power grid image positioning method and image positioning model training method |
| CN111062417A (en) * | 2019-11-19 | 2020-04-24 | 武汉大学 | Lie-Mean-based flat shell defect detection method and system |
| CN110942450A (en) * | 2019-11-19 | 2020-03-31 | 武汉大学 | Multi-production-line real-time defect detection method based on deep learning |
| CN111079629A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon center plate out-of-position fault image identification method |
| CN110969620A (en) * | 2019-12-19 | 2020-04-07 | 中国农业大学 | Method and device for detecting magnetic shoe ripple defects |
| CN110992361A (en) * | 2019-12-25 | 2020-04-10 | 创新奇智(成都)科技有限公司 | Engine fastener detection system and detection method based on cost balance |
| CN111369545A (en) * | 2020-03-10 | 2020-07-03 | 创新奇智(重庆)科技有限公司 | Edge defect detection method, device, model, equipment and readable storage medium |
| CN111369545B (en) * | 2020-03-10 | 2023-04-25 | 创新奇智(重庆)科技有限公司 | Edge defect detection method, device, model, equipment and readable storage medium |
| CN111400040A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Industrial Internet system based on deep learning and edge calculation and working method |
| CN111768404A (en) * | 2020-07-08 | 2020-10-13 | 北京滴普科技有限公司 | Mask appearance defect detection system, method and device and storage medium |
| CN111833328A (en) * | 2020-07-14 | 2020-10-27 | 汪俊 | Aircraft engine blade surface defect detection method based on deep learning |
| CN111833328B (en) * | 2020-07-14 | 2023-07-25 | 汪俊 | Aircraft engine blade surface defect detection method based on deep learning |
| CN112132196A (en) * | 2020-09-14 | 2020-12-25 | 中山大学 | Cigarette case defect identification method combining deep learning and image processing |
| CN112132196B (en) * | 2020-09-14 | 2023-10-20 | 中山大学 | Cigarette case defect identification method combining deep learning and image processing |
| CN112149594A (en) * | 2020-09-29 | 2020-12-29 | 同济大学 | Urban construction assessment method based on deep learning and high-resolution satellite images |
| CN112907519A (en) * | 2021-01-29 | 2021-06-04 | 广州信邦智能装备股份有限公司 | Metal curved surface defect analysis system and method based on deep learning |
| CN113112497A (en) * | 2021-05-06 | 2021-07-13 | 合肥中科迪宏自动化有限公司 | Industrial appearance defect detection method based on zero sample learning, electronic device and storage medium |
| CN113240647A (en) * | 2021-05-18 | 2021-08-10 | 哈尔滨工业大学（深圳） | Mobile phone shell rear cover defect detection method and system based on deep learning |
| CN113192061A (en) * | 2021-05-25 | 2021-07-30 | 安徽睿微信息科技有限公司 | LED package appearance detection image extraction method and device, electronic equipment and storage medium |
| CN113192061B (en) * | 2021-05-25 | 2023-08-29 | 安徽睿微信息科技有限公司 | Extraction method and device of LED package appearance detection image, electronic equipment and storage medium |
| CN113610754A (en) * | 2021-06-28 | 2021-11-05 | 浙江文谷科技有限公司 | Defect detection method and system based on Transformer |
| CN113610754B (en) * | 2021-06-28 | 2024-05-07 | 浙江文谷科技有限公司 | Defect detection method and system based on transducer |
| CN113705442A (en) * | 2021-10-09 | 2021-11-26 | 广东博媒广告传播有限公司 | Outdoor large-board advertising picture monitoring and identifying system and method |
| CN113744274B (en) * | 2021-11-08 | 2022-02-08 | 深圳市巨力方视觉技术有限公司 | Product appearance defect detection method and device and storage medium |
| CN113744274A (en) * | 2021-11-08 | 2021-12-03 | 深圳市巨力方视觉技术有限公司 | Product appearance defect detection method and device and storage medium |
| CN115147409A (en) * | 2022-08-30 | 2022-10-04 | 深圳市欣冠精密技术有限公司 | Mobile phone shell production quality detection method based on machine vision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106875381B (en) | 2020-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106875381A (en) | A kind of phone housing defect inspection method based on deep learning | |
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| CN110909690B (en) | Method for detecting occluded face image based on region generation | |
| CN110263705A (en) | Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method | |
| CN105069472B (en) | A kind of vehicle checking method adaptive based on convolutional neural networks | |
| CN109636772A (en) | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN107610087B (en) | Tongue coating automatic segmentation method based on deep learning | |
| CN109993094A (en) | Fault in material intelligent checking system and method based on machine vision | |
| CN109711474A (en) | A kind of aluminium material surface defects detection algorithm based on deep learning | |
| CN113592845A (en) | Defect detection method and device for battery coating and storage medium | |
| CN108009509A (en) | Vehicle target detection method | |
| CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
| CN107909081B (en) | Method for quickly acquiring and quickly calibrating image data set in deep learning | |
| CN111507426B (en) | Non-reference image quality grading evaluation method and device based on visual fusion characteristics | |
| CN108985170A (en) | Transmission line of electricity hanger recognition methods based on Three image difference and deep learning | |
| CN104077577A (en) | Trademark detection method based on convolutional neural network | |
| CN109145846A (en) | Material microstructure intelligent recognition analysis system and analysis method | |
| CN109376631A (en) | A kind of winding detection method and device neural network based | |
| CN108334955A (en) | Copy of ID Card detection method based on Faster-RCNN | |
| CN110082821A (en) | A kind of no label frame microseism signal detecting method and device | |
| CN109711416A (en) | Target identification method, device, computer equipment and storage medium | |
| CN108256462A (en) | A kind of demographic method in market monitor video | |
| CN108664839A (en) | A kind of image processing method and equipment | |
| CN106874825A (en) | The training method of Face datection, detection method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |